overthelex/ukrainian-court-decisions

Ukrainian Court Decisions — Judgment Prediction

A dataset of Ukrainian court decisions for case outcome prediction, extracted from the State Court Decisions Registry (ЄДРСР).

Task

Given the facts section (ВСТАНОВИВ) of a court decision, predict the judgment outcome:

Data Description

• —Source: State Court Decisions Registry of Ukraine (Єдиний державний реєстр судових рішень, ЄДРСР)
• —Language: Ukrainian (uk)
• —Jurisdiction: Ukraine
• —Court types: Civil and commercial courts
• —Decision form: "Рішення" (substantive decisions, judgment_code=3)
• —Time period: 2003–2025

Why facts only?

Following Niklaus et al. (2021), we provide only the facts section (ВСТАНОВИВ/УСТАНОВИВ) as input, excluding the court's reasoning and the dispositive (ruling) section. Including the full text would make the task trivially solvable since the ruling contains the answer directly.

Section extraction

Ukrainian court decisions follow a standard structure:

1. 1.Header — court name, case number, date, parties
2. 2.ВСТАНОВИВ/УСТАНОВИВ — established facts and circumstances
3. 3.Reasoning — court's legal analysis (sometimes merged with facts)
4. 4.ВИРІШИВ/УХВАЛИВ — the ruling (dispositive section)

We extract text between the ВСТАНОВИВ marker and the ВИРІШИВ marker. The outcome label is derived from the dispositive section using keyword matching (задовольнити → approved, відмовити → dismissed, частково → partial).

Data Quality

• —150,000 samples total (120K train / 15K validation / 15K test)
• —Perfectly balanced: 50,000 per class (approved / dismissed / partial)
• —Minimum 200 characters in facts section
• —Maximum 10,000 characters (truncated for very long decisions)
• —Full text length between 500 and 200,000 characters
• —Sampled from 596,538 valid extractions out of 802,189 court decisions
• —Random 80/10/10 train/validation/test split

Usage

from datasets import load_dataset

dataset = load_dataset("overthelex/ukrainian-court-decisions")

# Or load via LexTreme
dataset = load_dataset("joelniklaus/lextreme", "ukrainian_court_decisions_judgment")

Papers

This dataset is used in the following papers:

• —Tokenizer Fertility and Zero-Shot Performance of Foundation Models on Ukrainian Legal Text: A Comparative Study -- arXiv:2605.14890 [cs.CL]
• —Temporal Concept Drift in Legal Judgment Prediction: Neural Baselines Across Three Epochs of Ukrainian Court Decisions -- arXiv:2605.24452 [cs.CL]
• —Automatic Construction of a Legal Citation Graph from 100 Million Ukrainian Court Decisions: Large-Scale Extraction, Topological Analysis, and Ontology-Driven Clustering -- arXiv:2605.15362 [cs.CL]

Citation

@article{ovcharov2026fertility,
  title={Tokenizer Fertility and Zero-Shot Performance of Foundation Models on Ukrainian Legal Text: A Comparative Study},
  author={Ovcharov, Volodymyr},
  journal={arXiv preprint arXiv:2605.14890},
  year={2026},
  url={https://arxiv.org/abs/2605.14890}
}

@article{ovcharov2026temporaldrift,
  title={Temporal Concept Drift in Legal Judgment Prediction: Neural Baselines Across Three Epochs of Ukrainian Court Decisions},
  author={Ovcharov, Volodymyr},
  journal={arXiv preprint arXiv:2605.24452},
  year={2026},
  url={https://arxiv.org/abs/2605.24452}
}

@article{ovcharov2026citationgraph,
  title={Automatic Construction of a Legal Citation Graph from 100 Million Ukrainian Court Decisions: Large-Scale Extraction, Topological Analysis, and Ontology-Driven Clustering},
  author={Ovcharov, Volodymyr},
  journal={arXiv preprint arXiv:2605.15362},
  year={2026},
  url={https://arxiv.org/abs/2605.15362}
}

License

CC-BY-4.0. The source data is published by the State Court Administration of Ukraine under open access provisions per Ukrainian law.

Links

• —ЄДРСР (State Court Decisions Registry)
• —LexTreme Benchmark
• —SecondLayer Legal AI Platform